An adaptive feature mode decomposition method and application in fault identification of rolling bearing

Abstract

Surface injury is a kind of bearing failure that has strong mutability and high risk, whose early symptom is hard to be recognized. When this kind of fault occurs, vibration signals will often be accompanied by periodic impact. Feature mode decomposition (FMD) gives full consideration to the periodicity and impulsiveness of signals and thus has a significant advantage in extracting features of periodic faults. The signal period and configuration of the initial finite impulse response (FIR) filter bank determine the decomposition performance of FMD, but they are often difficult to be accurately determined. To solve these problems, the paper proposed an adaptive FMD (Adaptive-FMD) algorithm. The single autocorrelation of the original signal in FMD is replaced by the combination of quadruplicate envelope and dual autocorrelation to suppress noise and highlight periodic information, providing a guarantee for subsequent signal processing. The required period of FMD is accurately determined by the average delay difference of all adjacent extreme points after preprocessing with a second-order autocorrelation function instead of a single autocorrelation in the classic method and strategy of taking a delay difference of adjacent extreme points to determine a signal period. The feature frequency of a determined period is applied to FMD to adaptively determine the upper and lower limits of FIR filters, to address the randomness in manual adjustment of the number of FMD segmentation frequency bands when the sampling frequency of input signal changes. Besides, this setting can also solve the problem that FIR filters is completely independent of the determined signal period, resulting in a potential false judgment. The effectiveness and applicability of the proposed method are verified by analysis of simulation signals and signals of different testers in different fault types and rotation speeds according to the proposed method, and comparison with various typical methods, such as FMD and variational mode decomposition.